Factor analysis and prediction of startups and ways to exit based on decision tree classification models with adaptive k with SMOTE method for imbalance problem

Abstract

This paper focuses on factor analysis to combine the information of startups with an synthetic minority over-sampling technique (SMOTE) method via an aspect of the decision tree algorithms that assist investors in project screening for describing important features.However, the investment of a startup company has characteristics of imbalanced data. Improvements in the handling of imbalanced data based on the SMOTE method has been developed by sampling from the minority class. The problem is how to set optimized k-nearest neighbors among the most common feature values. This work purposed a method to fit data in the startup’s information that is designed to handle the data value by adaptive k with SMOTE, which manages the problem with an imbalance class label for robustness of evaluation metrics for balancing the portion of multi-class. The adaptive k experimental results can solve the k parameter setting and produce a high accuracy rate of startup companies’ class as closed, operating, and acquired status of investment at 0.84, 0.87 and 0.97 respectively. The overall accuracy rate is 0.99; that is the best outcome compared with other methods for handling imbalance. In addition, the results and discussion shown that can meet the needs of investment startup are designed and discussed of business views and machine learning views to work co-operation.